- title: 'Electrophysiologically Informed Neuromorphic Spiking Networks for Spatial Navigation' abstract: 'Spatial memory underlies the mental encoding, storage, and retrieval of spatial representations that support navigation in intricate environments. Conventional models of navigation accentuate the role of place and grid cells as principal neural substrates. However, recent findings in teleost fish, the most diverse vertebrate class, suggest that navigation in these species depends primarily on boundary vector cells (BVCs) and hydrostatic pressure (HP) cues. In this study, we designed a neuromorphic spiking neural network (SNN) for spatial navigation, directly informed by electrophysiological recordings from the goldfish telencephalon. Within this architecture, BVC populations mediated obstacle avoidance, while HP-sensitive units provided a vertical reference for goal-oriented trajectory planning. Our results demonstrate that efficient navigation can emerge without explicit positional coding, consistent with experimental observations of low firing rates and limited neuronal populations in the fish telencephalon. The proposed framework thus establishes a compact and biologically grounded model for fish-inspired neuromorphic navigation that remains robust and scalable across naturalistic conditions.' volume: 308 URL: https://proceedings.mlr.press/v308/cohen26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/cohen26a/cohen26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-cohen26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Lear family: Cohen - given: Hadar Cohen family: Duwek - given: Elishai Ezra family: Tsur editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 1-9 id: cohen26a issued: date-parts: - 2026 - 5 - 18 firstpage: 1 lastpage: 9 published: 2026-05-18 00:00:00 +0000 - title: 'Out-of-Distribution Generalization under Augmented Stimuli Reveals the Inductive Bias of Visual Cortex Digital Twins' abstract: 'An important goal in Neuro-AI is to develop a digital twin of the visual cortex. Currently, state-of-the-art models of the visual cortex require large amounts of training data, which are difficult to obtain for most neuroscience laboratories. Here, we propose an approach to alleviate this limitation by enhancing neuronal data quality through optimization of visual stimuli. We first evaluated various image-transformation methods in silico using CNN-based models of the mouse visual cortex. We then validated the selected methods in vivo using real mouse brain recordings. The in vivo experiments identified two methods that enhanced neuronal responses and accelerated the training of digital twins. Unexpectedly, one method (Sharpening) consistently failed to match the in silico predictions. This discrepancy was likely due to CNN’s inductive bias toward high spatial frequencies. Consistently, models of the visual cortex with more favorable spectral sensitivity successfully predicted in vivo neuronal responses to Sharpening-transformed images. Taken together, our work makes the following contributions toward the development of a digital twin of the visual cortex: 1) Two in vivo-validated image-transformation methods that enhance data quality and accelerate model training. 2) Evidence that the RNN-based model is more aligned with the real visual cortex than CNN- or ViT-based models.' volume: 308 URL: https://proceedings.mlr.press/v308/kasagi26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/kasagi26a/kasagi26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-kasagi26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Ayumi family: Kasagi - given: Takemi family: Hieda - given: Yuki family: Hosaka - given: Ruixiang family: Li - given: Masato family: Taki - given: Teppei family: Matsui editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 10-17 id: kasagi26a issued: date-parts: - 2026 - 5 - 18 firstpage: 10 lastpage: 17 published: 2026-05-18 00:00:00 +0000 - title: 'Multi-Modal Natural Intelligence through Active Predictive Coding' abstract: 'Active predictive coding (APC) is a recently proposed theory of the neocortex that postulates that a canonical sensory-motor processing circuit is replicated across cortical areas. These areas are organized in a rough hierarchy, with higher-level neural states modulating lower-level circuits implementing state-transition dynamics and policy functions. Such a structure enables the network to learn the compositional structure of the world, allowing it to rapidly compose solutions to new problems and generalize quickly to new environments. In APC, complex state transition dynamics are modeled as a sequence of simpler dynamics, which in turn are modeled using even simpler dynamics, and so on. Complex policies are similarly modeled as sequences of simpler policies, with the lowest level comprising sequences of primitive actions. Here we show that the APC model offers a unifying framework for multi-modal intelligence by demonstrating that the same architecture can (a) perform visual object recognition via active sensing (eye movements) and parts-based understanding, (b) navigate to desired goal locations in a complex environment through hierarchical planning, (c) learn to parse language hierarchically, infer the goal (i.e., intent) of an uttered sentence, and achieve the inferred goal through actions, and (d) scale up to realistic environments. Our results suggest that neurally-inspired approaches such as APC can help pave the way for more interpretable, generalizable, efficient, and human-like multi-modal AI.' volume: 308 URL: https://proceedings.mlr.press/v308/duan26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/duan26a/duan26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-duan26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Jeffrey family: Duan - given: Vishwas family: Sathish - given: Crimson family: Stambaugh - given: Rajesh P. N. family: Rao editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 18-26 id: duan26a issued: date-parts: - 2026 - 5 - 18 firstpage: 18 lastpage: 26 published: 2026-05-18 00:00:00 +0000 - title: 'DynaTab: Dynamic Feature Ordering as Neural Rewiring for High-Dimensional Tabular Data' abstract: 'High-dimensional tabular data lacks a natural feature order, limiting the applicability of permutation-sensitive deep learning models. We propose DynaTab, a dynamic feature ordering-enabled architecture inspired by neural rewiring. We introduce a lightweight criterion that predicts when feature permutation will benefit a dataset by quantifying its intrinsic complexity. DynaTab dynamically reorders features via a neural rewiring algorithm and processes them through a compact, dynamic order-aware combination of separate learned positional embedding, importance-based gating, and masked attention layers, compatible with any sequence-sensitive backbone. Trained end-to-end with bespoke dynamic feature ordering (DFO) and dispersion losses, DynaTab achieves statistically significant gains, particularly on high-dimensional datasets, where it is benchmarked against 45 state-of-the-art baselines across 36 different real-world tabular datasets. Our results position DynaTab as a compelling new paradigm for high-dimensional tabular deep learning.' volume: 308 URL: https://proceedings.mlr.press/v308/habib26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/habib26a/habib26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-habib26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Al Zadid Sultan Bin family: Habib - given: Gianfranco family: Doretto - given: Donald A. family: Adjeroh editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 27-57 id: habib26a issued: date-parts: - 2026 - 5 - 18 firstpage: 27 lastpage: 57 published: 2026-05-18 00:00:00 +0000 - title: 'Developing Autoencoder: Incremental Bottleneck Expansion Leads to an Informed Latent Space' abstract: 'Representation learning models, such as autoencoders (AEs), can effectively extract meaningful and generalizable features from natural image data. However, the learned latent features are often mixed or distributed across all bottleneck units, making interpretation difficult. Previous work has sought to address this by explicitly optimizing for feature separation or ordering. We propose a biologically inspired progressive learning scheme, the Developing Autoencoder (Dev-AE), which incrementally expands the representational capacity. Increasing the size of the bottleneck layer over training epochs forces the Dev-AE to first learn compressed, low-dimensional representations before expanding into progressively higher-dimensional feature spaces. Comparing the latent space organization in Dev-AEs with that in standard AEs and PCA-initialized AEs (PCA-AE), we observe improved feature ordering and higher activation sparsity. Moreover, Dev-AEs show better classification performance based on the learned encodings, with units added in the final increment contributing the most. Our findings indicate that an incremental latent space expansion fosters ordered, sparse, and more diverse representations, leading to more efficient use of representational capacity and improved classification accuracy, thereby offering a promising route toward interpretable and compact encodings.' volume: 308 URL: https://proceedings.mlr.press/v308/kong26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/kong26a/kong26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-kong26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Deyue family: Kong - given: Jonas family: Elpelt - given: David family: Vogenauer - given: Markos family: Genios - given: Matthias family: Kaschube editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 58-66 id: kong26a issued: date-parts: - 2026 - 5 - 18 firstpage: 58 lastpage: 66 published: 2026-05-18 00:00:00 +0000 - title: 'A Biologically Interpretable Cognitive Architecture for Online Structuring of Episodic Memories into Cognitive Maps' abstract: 'Cognitive maps provide a powerful framework for understanding spatial and abstract reasoning in biological and artificial agents. While recent computational models link cognitive maps to hippocampal-entorhinal mechanisms, they often rely on global optimization rules (e.g., backpropagation) that lack biological plausibility. In this work, we propose a novel cognitive architecture for structuring episodic memories into cognitive maps compatible with neural substrate constraints. Our model integrates the Successor Features framework with episodic memories, enabling incremental, online learning through agent-environment interaction. We demonstrate its efficacy in a partially observable gridworld, where the architecture autonomously organizes memories into structured representations without centralized optimization. This work bridges computational neuroscience and AI, offering a biologically grounded approach to cognitive map formation in artificial adaptive agents.' volume: 308 URL: https://proceedings.mlr.press/v308/dzhivelikian26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/dzhivelikian26a/dzhivelikian26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-dzhivelikian26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Evgenii family: Dzhivelikian - given: Nikita family: Bainaev-Mangilev - given: Aleksandr family: Panov editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 67-75 id: dzhivelikian26a issued: date-parts: - 2026 - 5 - 18 firstpage: 67 lastpage: 75 published: 2026-05-18 00:00:00 +0000 - title: 'Hierarchical Predictive Processing for Uncertainty-Aware Multimodal Transformers' abstract: 'Current vision-language models suffer from overconfident predictions and cross-modal hallucinations, lacking principled mechanisms for uncertainty quantification. We introduce a novel architecture that applies the Free Energy Principle from computational neuroscience to multimodal transformers, enabling reliable uncertainty estimation through hierarchical predictive processing. Our approach implements precision-weighted cross-modal prediction, where visual and linguistic representations generate predictions about each other, and prediction errors are weighted by learned precision matrices that capture cross-modal consistency. By minimizing variational free energy across modalities, our model naturally quantifies uncertainty while maintaining task performance. Experimental results demonstrate substantial improvements over standard uncertainty quantification methods, achieving 51.7% better calibration than Monte Carlo Dropout baselines on synthetic evaluation data and 48.6% improvement on the VQA v2 dataset. This work establishes the first principled bridge between the brain’s Bayesian inference mechanisms and practical multimodal AI uncertainty quantification, demonstrating that biologically-inspired architectures can significantly enhance model reliability.' volume: 308 URL: https://proceedings.mlr.press/v308/achyuthan26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/achyuthan26a/achyuthan26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-achyuthan26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Namita family: Achyuthan - given: Bhaskarjyoti family: Das editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 76-83 id: achyuthan26a issued: date-parts: - 2026 - 5 - 18 firstpage: 76 lastpage: 83 published: 2026-05-18 00:00:00 +0000 - title: 'CATFormer: When Continual Learning Meets Spiking Transformers With Dynamic Thresholds' abstract: 'Although deep neural networks perform extremely well in controlled environments, they fail in real-world scenarios where the data isn’t available all at once, and the model requires an update to adapt itself to the new data distribution, which might or might not follow the initial distribution. Previously acquired knowledge is lost during such subsequent updates from new data. a phenomenon commonly known as catastrophic forgetting. In contrast, the brain can learn without such catastrophic forgetting, irrespective of the number of tasks it encounters. Existing spiking neural networks (SNNs) for class-incremental learning (CIL) suffer a sharp performance drop as tasks accumulate. We here introduce CATFormer (Context Adaptive Threshold Transformer), a scalable framework that overcomes this limitation. We observe that the key to preventing forgetting in SNNs lies not only in synaptic plasticity, but in modulating neuronal excitability too. At the core of CATFormer is the Dynamic Threshold Leaky Integrate-and-Fire (DTLIF) neuron model, which leverages context-adaptive thresholds as the primary mechanism for knowledge retention. This is paired with a Gated Dynamic Head Selection (G-DHS) mechanism for task-agnostic inference. Extensive evaluation on both static (CIFAR-10/100/Tiny-ImageNet) and neuromorphic (CIFAR10-DVS/SHD) datasets reveals that CATFormer outperforms existing rehearsal-free CIL algorithms across various task splits, establishing it as an ideal architecture for energy-efficient and true class incremental learning.' volume: 308 URL: https://proceedings.mlr.press/v308/nagabhushana26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/nagabhushana26a/nagabhushana26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-nagabhushana26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Vaishnavi family: Nagabhushana - given: Kartikay family: Agrawal - given: Ayon family: Borthakur editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 84-92 id: nagabhushana26a issued: date-parts: - 2026 - 5 - 18 firstpage: 84 lastpage: 92 published: 2026-05-18 00:00:00 +0000 - title: 'Masked Autoencoders Learn Perception-Relevant Representations from Resting State Neural Data' abstract: 'Clinical neuroprosthetics face a data bottleneck: labeled perception trials are scarce while hours of spontaneous neural activity are largely underutilized. Here, we test whether self-supervised learning can use these unlabeled datasets to improve perception decoding. We pretrained a masked autoencoder on 14.6 hours of spontaneous multiunit activity from an intracortical array in a blind participant’s V1. The model captured interpretable brain structure without supervision: V1’s spatial organization and perceptual state separation both emerged purely from its latent representations. To test these features, we used linear probing (logistic regression on the frozen latents) to measure performance on the data with stimulation. Perception decoding accuracy reached 84.1% on a general psychometric task. On the more difficult threshold-level task, accuracy reached 64.0%. This work shows that spontaneous cortical activity is not noise; it contains rich, task-relevant structure. Unsupervised pretraining on this data is a promising strategy to improve neural decoding.' volume: 308 URL: https://proceedings.mlr.press/v308/kovalev26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/kovalev26a/kovalev26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-kovalev26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Aleksandr family: Kovalev - given: Antonio family: Lozano - given: Fabrizio family: Grani - given: Cristina family: Soto Sanchez - given: Leili family: Soo - given: Rocío family: López-Peco - given: Adrian family: Villamarin-Ortiz - given: Roberto family: Morollón Ruiz - given: María del Mar family: Ayuso Arroyave - given: Alfonso family: Rodil - given: Eduardo family: Fernández editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 93-98 id: kovalev26a issued: date-parts: - 2026 - 5 - 18 firstpage: 93 lastpage: 98 published: 2026-05-18 00:00:00 +0000 - title: 'Play the (Mis)Match: Using fMRI-Aligned Feature Fine-Tuning to Reveal Shortcut Bias in Deep Neural Networks' abstract: 'Deep neural networks (DNNs) often “cheat” by relying on shortcut objects (e.g., food$\Rightarrow$kitchen) rather than holistic spatial layout, undermining out-of-distribution (OOD) robustness. This work serves as a proof-of-concept exploration of whether fMRI alignment can reduce shortcut bias in visual DNNs. We address this issue with Play the (Mis)Match, a diagnostic dataset and brain-aligned fine-tuning framework. Leveraging fMRI recordings from the Natural Scenes Dataset (four participants; bedroom, bathroom, living room, kitchen), we curate MATCH images in which shortcut cues co-occur as usual and MISMATCH images from which those cues are removed. ImageNet-initialised CNN and Transformer backbones are fine-tuned with an MSE alignment loss that steers their intermediate features toward voxel patterns known to be less sensitive to shortcut cues. Our results show that, for ResNet, this procedure narrows the Match–Mismatch accuracy gap by 24 % and redirects Grad-CAM attention from individual objects to holistic scene structure, particularly activity from the scene-selective cortex (PPA, RSC, OPA), all without explicit shortcut annotations. Our study provides a proof-of-concept that human-brain constraints may help steer DNNs toward more semantically grounded, less shortcut-dependent scene representations.' volume: 308 URL: https://proceedings.mlr.press/v308/lin26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/lin26a/lin26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-lin26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Yang Chen family: Lin - given: Chiayun family: Lee - given: Po-Chih family: Kuo editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 99-107 id: lin26a issued: date-parts: - 2026 - 5 - 18 firstpage: 99 lastpage: 107 published: 2026-05-18 00:00:00 +0000 - title: 'Similar Accuracy but Different Topographies under Cross-Entropy and Contrastive Learning' abstract: 'The brain’s topographic organization has motivated topographic deep neural networks (TDNNs) as models of perceptual and conceptual representation. However, prior TDNN studies largely paired topography with cross-entropy (CE). They have not examined whether contrastive objectives are generally compatible with topographic training, how topographic strength affects run-to-run representational consistency, or what failure modes limit the effect of the topographic constraint. We addressed these issues by training TDNNs on CIFAR-10 with a local topographic loss that minimized the average l2 distance between afferent weight vectors of neighboring units. We compared four objectives: CE, supervised contrastive, self-supervised SimCLR, and a label-aware contrastive margin loss reflecting an animacy hierarchy. Across topographic strengths, label-supervised objectives maintained high accuracy, produced smooth activation maps, and increased within-class similarity relative to CE. Two factors limited the impact of the topographic loss: 1) dropout was required to obtain smooth maps rather than sparse activations; 2) under strong penalties, networks reduced the topographic loss by shrinking weight norms rather than aligning weight directions. We also found that stronger topographic constraints reduced cross-seed representational consistency, indicating multiple comparably good topographic solutions. Nonetheless, ensembles built from sets of less-consistent models only slightly outperformed ensembles without topographic constraints. Our results indicate that contrastive objectives are a robust option for training topographic networks, producing good accuracy and high within-class similarity. The findings also identify boundary conditions for afferent-weight similarity as a topographic prior.' volume: 308 URL: https://proceedings.mlr.press/v308/sander26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/sander26a/sander26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-sander26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Gerrit family: Sander - given: Uri family: Hasson editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 108-115 id: sander26a issued: date-parts: - 2026 - 5 - 18 firstpage: 108 lastpage: 115 published: 2026-05-18 00:00:00 +0000 - title: 'Decoding Cortical Microcircuits: A Generative Model for Latent Space Exploration and Controlled Synthesis' abstract: 'A central idea in understanding brains and building brain neural networks is that structure determines function. However, the brain’s connectome is a massively high-dimensional graph, making the direct investigation of its structure-function relationships computationally intractable. Therefore, identifying a compact, low-dimensional representation that captures the connectome’s essential structural organization is crucial for elucidating these relationships. The existence of such a representation is biologically plausible: the "genomic bottleneck" theory provides a strong basis for such a compressed developmental blueprint. We introduce a generative model to learn this underlying representation from detailed connectivity maps of mouse cortical microcircuits. Our model successfully captures the essential structural information of these circuits within a compressed latent space. We then associate specific network structures, as encoded in this latent space, with computational functions using reservoir computing tasks. Building on this, our methodology allows for the controllable generation of novel, synthetic microcircuits with desired structural features by navigating the learned latent space. This research paradigm establishes a computational testbed to systematically investigate the brain’s inherent structure-function relationships. The ability to generate diverse, bio-plausible circuits could inform the development of more advanced artificial neural networks.' volume: 308 URL: https://proceedings.mlr.press/v308/liu26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/liu26a/liu26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-liu26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Xingyu family: Liu - given: Yubin family: Li - given: Guozhang family: Chen editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 116-148 id: liu26a issued: date-parts: - 2026 - 5 - 18 firstpage: 116 lastpage: 148 published: 2026-05-18 00:00:00 +0000 - title: 'Building Emotional Intelligence into Digital Therapy AI Agents through Neurofeedback' abstract: 'We present a novel emotionally intelligent agent framework for delivering cognitive behavioural therapy (CBT). The system aggregates text sentiment cues with neurofeedback, yielding a fine-grained perception of user state building empathy into the agent. A reinforcement learning (RL) planner maps this affective state to appropriate therapeutic acts, which are verbalised by a large language model (LLM). To enhance reliability, the LLM agent is augmented with a meta-cognitive control layer that continuously self-monitors and refines of its responses. In preliminary studies, the proposed system has demonstrated improved therapeutic efficacy over standard LLM-based agents, as measured by standard psychotherapy metrics. These results highlight the potential of combining neurofeedback, affective computing, RL decision making, and LLM generation to deliver clinically meaningful, scalable CBT paving the way for safe, personalised mental health support at population scale.' volume: 308 URL: https://proceedings.mlr.press/v308/nallaperuma-herzberg26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/nallaperuma-herzberg26a/nallaperuma-herzberg26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-nallaperuma-herzberg26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Sam family: Nallaperuma-Herzberg - given: Rishabh family: Balse - given: Sonia family: Koszut - given: Lilith family: Stenhouse - given: Anna family: Bevan - given: Tristan family: Bekinschtein - given: Pietro family: Lio editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 149-154 id: nallaperuma-herzberg26a issued: date-parts: - 2026 - 5 - 18 firstpage: 149 lastpage: 154 published: 2026-05-18 00:00:00 +0000 - title: 'MultiDiffNet: A Multi-Objective Diffusion Framework for Generalizable Brain Decoding' abstract: 'Neural decoding from electroencephalography (EEG) remains fundamentally limited by poor generalization to unseen subjects, driven by high inter-subject variability and the lack of large-scale datasets to model it effectively. Existing methods often rely on synthetic subject generation or simplistic data augmentation, but these strategies fail to scale or generalize reliably. We introduce MultiDiffNet, a diffusion-based framework that bypasses generative augmentation entirely by learning a compact latent space optimized for multiple objectives. We decode directly from this space and achieve state-of-the-art generalization across various neural decoding tasks using subject and session disjoint evaluation. We also curate and release a unified benchmark suite spanning four EEG decoding tasks of increasing complexity (SSVEP, Motor Imagery, P300, and Imagined Speech) and an evaluation protocol that addresses inconsistent split practices in prior EEG research. Finally, we develop a statistical reporting framework tailored for low-trial EEG settings. Our work provides a reproducible and open-source foundation for subject-agnostic EEG decoding in real-world BCI systems.' volume: 308 URL: https://proceedings.mlr.press/v308/zhang26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/zhang26a/zhang26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-zhang26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Meng-Chun family: Zhang - given: Kateryna family: Shapovalenko - given: Yucheng family: Shao - given: Eddie family: Guo - given: Parusha family: Pradhan editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 155-162 id: zhang26a issued: date-parts: - 2026 - 5 - 18 firstpage: 155 lastpage: 162 published: 2026-05-18 00:00:00 +0000 - title: 'Neural Attention Maps Alignment in Vision Transformers and Mammalian Visual Cortex' abstract: 'Image parsing with Vision Transformers has achieved state-of-the-art results, but how these models process visual information compared to biological vision systems is an open question. In this study, we present an extensive benchmarking between the attention mechanisms in the Vision Transformer-based models, such as Segment Anything, and its several variants that capture long-range dependencies in understanding the generalized features in natural images, with the neural responses captured from the mouse visual cortex for the same visual inputs. We found a significant correspondence between self-attention and convolutional maps in these models and cortical neural activity in the mouse visual cortex. This trend is observed to be consistent across similar model architectures with varying numbers of parameter units and provides an explainable trade-off between the accuracy and efficiency on real-world object segmentation datasets. This relationship is observed to be generalized across the sub-regions and neuronal genotypes, capturing diverse functional units in the mouse visual cortex. Our work proposes a pioneering effort in identifying important parallels between hierarchical representational learning in vision-based transformers and the biological visual cortex. To advance the development of neuro-AI models, these neural correlates suggest that aspects of cortical computation, captured by the state-of-the-art vision models, can potentially contribute to their effectiveness for image understanding tasks as well as guiding the advancement of novel model architecture design. We anticipate that this practice will also lead to future interpretability work to better understand the encoding and decoding principles of computation in the mammalian visual cortex.' volume: 308 URL: https://proceedings.mlr.press/v308/jalil26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/jalil26a/jalil26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-jalil26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Hamd family: Jalil - given: Ahmed Rashid family: Qazi - given: Asim family: Iqbal editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 163-179 id: jalil26a issued: date-parts: - 2026 - 5 - 18 firstpage: 163 lastpage: 179 published: 2026-05-18 00:00:00 +0000 - title: 'G-LaD: Graph-Language alignment for few-shot Diagnosis from fMRI' abstract: 'Decoding human cognitive states from neural activity is a core challenge in artificial intelligence and computational neuroscience. Functional Magnetic Resonance Imaging (fMRI) captures high-dimensional spatiotemporal patterns of brain activity, yet characterizing cognitive states based on modeling the complex, dynamic dependencies among distributed regions remains difficult. While Graph Neural Networks (GNNs) to represent the brain as a structured graph has advanced functional connectivity (FC) analysis, they suffer from limited generalization, reliance on large labeled datasets, and poor transferability across neuro-imaging tasks. We introduce Graph-Language alignment for Diagnosis (G-LaD), that integrates graph representation learning with Large Language Models (LLMs) for data-efficient brain graph classification. G-LaD first pretrains a graph encoder, built upon Graph Isomorphism Network layers using a reconstruction-driven Denoising Autoencoder, to capture structural and topological invariants. In the second stage, distribution-level alignment between graph and language representations is achieved via a Sinkhorn-divergence objective, enabling smooth and transferable cross-modal mapping. Finally, a Chain-of-Thought prompting mechanism guides the LLM to perform reasoning-driven predictions. Empirical evaluations on the ABIDE dataset demonstrate superior few-shot generalization and robust performance of G-LAD in neuro-degenerative disorder classification.' volume: 308 URL: https://proceedings.mlr.press/v308/gupta26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/gupta26a/gupta26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-gupta26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Abhishek family: Gupta - given: Vipul Kumar family: Singh - given: Jyotismita family: Barman - given: Sandeep family: Kumar - given: Anish family: Arora editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 180-188 id: gupta26a issued: date-parts: - 2026 - 5 - 18 firstpage: 180 lastpage: 188 published: 2026-05-18 00:00:00 +0000 - title: 'BRAINS: Building Representations with Autoencoders for Individualized Neuroimaging Spaces' abstract: 'Functional Magnetic Resonance Imaging (fMRI) provides rich, high-dimensional data on human brain activity, yet traditional dimensionality-reduction techniques primarily capture group-level structure and overlook individual variability. We introduce BRAINS, a framework based on Convolutional Variational Autoencoders (CVAEs) that learns subject-specific latent spaces directly from BOLD signals. These latent representations effectively denoise voxel-wise time series ( 5% tSNR gain) while preserving functional connectivity and anatomical coherence. Using Procrustes alignment, we show that individual latent spaces can be aligned across participants, revealing both shared and idiosyncratic components of cortical organization. Our approach bridges neuroimaging and deep representation learning, offering a geometry-aware foundation for individualized brain analysis and multimodal integration across subjects, tasks, and models. The code is available at: https://github.com/neural-data-science-lab/NEUROAI_AAAI_BRAINS.git.' volume: 308 URL: https://proceedings.mlr.press/v308/singla26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/singla26a/singla26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-singla26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Kajal family: Singla - given: Pierre-Louis family: Bazin - given: Nico family: Scherf editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 189-198 id: singla26a issued: date-parts: - 2026 - 5 - 18 firstpage: 189 lastpage: 198 published: 2026-05-18 00:00:00 +0000 - title: 'Steering Transformer Attention with Human EEG' abstract: 'Modern LLMs differ fundamentally from the human brain in architecture and computational mechanisms, yet recent work reveals surprising representational alignments between these systems. Here we test whether noninvasive neural signals can directly steer transformer attention at inference time. Using an InstABoost-style framework, we inject EEG-derived attention weights (suppressing alpha, enhancing theta/gamma bands) into early layers of Llama-3.2-3B without additional training. On reading comprehension tasks from the ZuCo dataset, we observe modest but consistent improvements (0.4-1.4% absolute gain), particularly when using population-averaged EEG from task-specific reading conditions. Control experiments with shuffled or misaligned EEG confirm these gains stem from temporal alignment between neural dynamics and word sequences. While preliminary, these results suggest that human attentional rhythms encode routing information that can productively guide artificial attention mechanisms, opening possibilities for neural-augmented language models.' volume: 308 URL: https://proceedings.mlr.press/v308/short26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/short26a/short26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-short26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Claire family: Short - given: Steven family: Basart - given: Sinem family: Erisken editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 199-204 id: short26a issued: date-parts: - 2026 - 5 - 18 firstpage: 199 lastpage: 204 published: 2026-05-18 00:00:00 +0000 - title: 'Shared Latent Coordinates and Multi-Observable Phase-Diagram Placement Yield Directly Comparable Mechanistic Fingerprints of Whole-Brain Dynamics' abstract: 'We present a methods-first framework that turns high-dimensional population neural recordings into directly comparable, mechanistic fingerprints at the level of individual subjects. Our pipeline (i) constructs population-universal, shared latent coordinates that align heterogeneous subjects into a common representational space; (ii) fits pairwise maximum-entropy (Ising) models on binarised latent trajectories with rigorous convergence- and uncertainty-diagnostics; (iii) performs energy-landscape analysis (ELA) to obtain interpretable minima, barriers and kinetic descriptors; and (iv) introduces a new, variance-balanced multi-observable phase-diagram analysis (PDA) that places many subjects - including systematically heterogeneous sub-groups - onto a shared Sherrington-Kirkpatrick (SK) reference surface with uncertainty, making cross-subject comparisons direct and faithful. In a cohort of rodent whole-brain imaging time series (sensitive third-party data), our placement costs are typically 10e-6 - 10e-4, with tight bootstrap confidence regions and consistent ordering across pooled and subgroup references; estimated SK parameters fall in s = 0.155-0.320, u = -0.013 to +0.031. The result is a compact, uncertainty-aware subject “fingerprint” comprising ELA and kinetic descriptors together with the subject’s location on the phase diagram. This paper focuses on methodological reliability and cross-subject comparability; external replications on public datasets remain future work.' volume: 308 URL: https://proceedings.mlr.press/v308/kedys26a.html PDF: https://raw.githubusercontent.com/mlresearch/v308/main/assets/kedys26a/kedys26a.pdf edit: https://github.com/mlresearch//v308/edit/gh-pages/_posts/2026-05-18-kedys26a.md series: 'Proceedings of Machine Learning Research' container-title: 'Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026' publisher: 'PMLR' author: - given: Julian family: Kedys - given: Cezary family: Mazurek editor: - given: Reza family: Abbasi-Asl - given: Asim family: Iqbal - given: Shinya family: Ito - given: Anton family: Arkhipov - given: Sophia family: Sanborn page: 205-213 id: kedys26a issued: date-parts: - 2026 - 5 - 18 firstpage: 205 lastpage: 213 published: 2026-05-18 00:00:00 +0000